Current convention

One PICO for questions about medical tests has been advocated by many: Population, Index test, Comparator test and test accuracy as Outcome.2–4 The index test is the new diagnostic or screening test under investigation, and the comparator test is the best available (reference) method for diagnosing the disease of interest. Although this PICO has strong face validity, it is ambiguous when studied in more detail. The index test as well as comparator test can be positive and negative and so it is unclear which results are compared. Additionally, the proposed outcome is not a health state but a statistical measure, such as sensitivity and specificity.

Moreover, the aim of some diagnostic research is to investigate the effect that offering a test has on patient health instead of test accuracy.5 This research acknowledges that the test result will determine which care the patient gets, and this will influence his health. It has been suggested that a PICO for a medical test should represent this type of research.6 7

Finally, although reviews about interventions are generally guided by an objective in terms of a PICO, reviews about test accuracy are not. Cochrane recommends to describe the objective of such a review as follows: ‘To determine the diagnostic accuracy of [index test] for detecting [target condition] in [participant description]’.8 This is obviously a clear and concise formulation, but the use of a PICO would promote consistency across reviews.

PICO for test accuracy

Let’s say we want to know whether the result of a mammography (X-ray) reflects the presence of breast cancer accurately in women aged 40 years or older without symptoms. We could phrase this clinical question as follows: is the risk of having breast cancer higher in women with a positive mammography compared with women with a negative mammography? If so, how strong is the predictive strength of the test? The PICO for this and similar questions would be: P is the population of interest, I stands for positive index test result, C for negative index test result and O for the target disease (box 1).

We would need to perform a test accuracy study to answer our question. The participants would get a mammography (index test) and a biopsy to determine the presence of breast cancer (reference test). Each test would yield a positive or negative test result. Preferably, all patients get both tests, without much time in between, and without the test interpreters knowing the result of the other test. However, to avoid subjecting a large group of women to an invasive biopsy, only women with a positive mammography get a biopsy. Cases in women with a negative mammography are detected as part of usual care, which includes a biopsy during 1 year of follow-up after the screening.

After finishing the study, we could present how many patients had breast cancer according to the biopsy (in the columns) by mammography result (in the rows). The figures in box 1 stem from the baseline measurement of the Swedish Two-County trial.9 10 Sensitivity and specificity were high (>95%), but since the prevalence of breast cancer was low (0.6%), the positive predictive value was very low (12.0%) and the negative predictive value very high (99.97%). In other words, the answer to our diagnostic question is that mammography picked up most cases of breast cancer but produced many false-positive results too.

PICO for care with test

What we want to know next is whether screening with mammography improves women’s health. Are they better off if we provide this type of care? The outcome all-cause death would capture the targeted benefit (preventing death due to breast cancer) and possible fatal adverse events (due to repeated radiation or overdiagnosis leading to unnecessary treatment for breast cancer). If we set up a study, we would compare care with screening to care without screening. Hence, the PICO is composed in the same way as a PICO for a question about any other type of intervention (box 2).

We would need to perform an intervention study, preferably a randomised trial. Women aged 40 years or older without symptoms would be randomised to care with mammography or to usual care without mammography, and deaths in both groups monitored. As the yearly breast cancer incidence is low (<1%), a large number of women and long follow-up that spans many years would be needed to accrue sufficient events and so adequate power.

We could use a 2×2 table to present how many women died (in the columns) by intervention group (in the rows). The numbers in the box 2 are based on the Swedish Two-County trial that tested the effect of screening.11 Of the 77 080 women screened every 2–3 years during 10 years, 7261 died (9.4% in total: 0.2% from breast cancer, 9.2% from other disease). Among the 55 985 women who were not screened in that same period, 5252 died (9.4% in total: 0.3% from breast cancer, 9.3% from other disease). Hence, the answer to our clinical question is that women did not benefit from mammographic screening in terms of the risk of dying.

Final remarks

In this paper, I distinguished two types of clinical questions about a medical test: is the test accurate, and does the use of the test improve patient health effectively? When teaching evidence-based medicine, it is important to make the distinction for several reasons. First, the nature of the two types of questions differs. A clinical question about a test’s accuracy relates to the accurate measurement of a health outcome: does the test result indicate the presence of the target disease good enough for a doctor to rely on it? In contrast, a question about an intervention based on such a test, such as, a screening programme, is about the intervention's effect on health: does the intervention change the risk of a certain health outcome in the participants?

Second, the type of question determines how the PICO is formulated. I explained the options above. Alternatively, some teachers might prefer to use just the one PICO structure that the two PICOs have in common on a meta-level: P stands for the population of interest, I for the investigated condition, C for the comparison condition and O for a health outcome. The choice is up to the teacher.

Third, the type of question determines which type of study answers the question. A test accuracy study compares the presence of the target disease in patients with a positive versus negative result on the index test cross-sectionally. An intervention study compares the health of patients who received care including the test with patients who did not receive this care during a certain period of follow-up. It could be an observational or randomised study.

Finally, the distinction between the two diagnostic questions is important, because high test accuracy does not automatically imply a health benefit to patients undergoing the test. The results of Swedish Two-County trial that I cited in the examples illustrated this point, and these findings have been corroborated by many other studies about breast cancer screening.12–15 Nevertheless, most diagnostic studies are designed to evaluate the test accuracy of a single test. More studies are needed that investigate the test accuracy of the customised diagnostic pathway with the new test, that is, the test strategy, or the effect that care based on this pathway has on health.6 16